With the continuous development of IoT technology, edge computing architecture is receiving more and more attention. Compared to traditional cloud computing architecture, edge computing architecture distributes data processing and storage closer to IoT terminal devices, effectively reducing data latency during transmission and reducing dependence on cloud resources. With the increasing number of IoT devices, more and more devices are accessing through the edge computing gateway, and more network requests need to be load balanced at the edge computing gateway for processing by the edge computing cluster. This article focuses on researching the issue of improving the access and load balancing capabilities of the edge computing gateway, with the following main research achievements:
(1) In response to the limitation of the performance of traditional Linux kernel protocol stack in processing network packets, optimization was carried out from the software level using DPDK technology. A user space protocol stack based on DPDK technology was designed due to the incomplete protocol stack provided by DPDK technology. The protocol stack first directly retrieves data from the network card and stores it in user space through polling, and uses a lock-free queue to pass data and synchronize the order between the data and control modules. Secondly, by setting CPU affinity for each module, internal data processing performance loss of the protocol stack was reduced. Finally, a multi-I/O multiplexing EPOLL interface method was designed to improve the ease of porting various Linux network applications to the protocol stack. This achieved efficient and easily portable access capability improvement for the edge computing gateway.
(2) In response to the performance defects of the centralized polling traversal of dynamic weighted load balancing algorithm for edge computing node schemes, a distributed dynamic weighted load balancing method based on DPDK was proposed. This method uses the DPDK user space protocol stack to accelerate the data processing capability of the edge computing gateway and edge computing nodes, and the edge computing cluster nodes push load weights as needed to reduce the bandwidth and performance loss of the edge computing gateway. By using the information entropy-based method to quickly detect fluctuations in edge computing node computing resources, a threshold support was provided for the push strategy of the edge computing node, enabling the load balancing strategy of the edge computing gateway to accurately match the computing resource situation of the edge computing node. This achieved an improvement in the load balancing capabilities of the edge computing gateway.
(3) A DPDK protocol stack-based access and load balancing system was designed and implemented. Performance tests show that the proposed improvement methods are reasonable and effective, and can effectively improve the network packet processing capability of the edge computing layer.
{
// console.log("22222");
// resolve(res);
..." />
